XiaoSong9905/Deep-Painterly-Harmonization-in-PyTorch

This PyTorch implementation follow the structure of Neural Style Pt Github Link by Justin Johnson where the network is first build and feature map is captured after the architrcture is build. In the original code Official Code Github Link, the feature map is captured during the build of architecture which cause waist of computation. Also, the loss in different layer back prop by simply adding them up and call loss_total.backward() For more information on how to specify training process, check main.py - get_args()